Topic：Sparse Optimization for Motion Segmentation

Time：2014年11月5日（周三）下午 2:00 - 3:00

Venue：信电大楼-215学术厅

Speaker：Dr. Michael Yang,

                  Leibniz University Hanover, Germany

Biography

Dr. Michael Yang is currently a postdoctoral research associate at the Institute for Information processing, Leibniz University Hanover, Germany. He received his Ph.D. from University of Bonn in 2011. His research interests are in the areas of computer vision, with focuses on probabilistic graphical models, multi-sensor fusion, scene understanding, and object segmentation. He serves as a regular reviewer for major journals and program committee member. He organized several workshops, e.g. CVPR Workshop on Multi-sensor Fusion, ECCV Workshop on Graphical Models.

Abstract

Motion segmentation aims to decompose a video sequence into different moving objects that move throughout the sequence. In this talk, I will show some state-of-the-art motion segmentation methods and our new framework based on subspace clustering with sparse optimization. We combine two sparse representations to optimize both the global and local estimation. Sparse PCA is applied for the global optimization. The local subspace separation is achieved via automatically selecting the sparse nearest neighbours. In the end of this talk, I will show some experimental results on the Hopkins 155 Dataset and Freiburg- Berkeley Dataset.